[BUG] M306 T produces negative responsiveness value

[Drzet]

Did you test the latest bugfix-2.1.x code?

Yes, and the problem still exists.

Bug Description

Having installed a new, original E3D Revo 12v hotend I ran M306 and got the below values:

:Recv: MPC Autotune finished! Put the constants below into Configuration.h Recv: MPC_BLOCK_HEAT_CAPACITY 9.36 Recv: **MPC_SENSOR_RESPONSIVENESS -0.9229** Recv: MPC_AMBIENT_XFER_COEFF 0.0525 Recv: MPC_AMBIENT_XFER_COEFF_FAN255 0.0884

The negative responsiveness value led to a thermal runaway issue until I realized what was happening and removed the minus sign.

Bug Timeline

With a new E3D Revo.

Expected behavior

Normal values

Actual behavior

Recv: MPC_SENSOR_RESPONSIVENESS -0.9229

Steps to Reproduce

T1 (it's a dual system)
M306 T

Version of Marlin Firmware

2.1.2.3

Printer model

Custom coreXY Hypercube

Electronics

SKR 3

LCD/Controller

BTT-12864

Other add-ons

No response

Bed Leveling

UBL Bilinear mesh

Your Slicer

Prusa Slicer

Host Software

OctoPrint

Don't forget to include

• A ZIP file containing your Configuration.h and Configuration_adv.h.

Additional information & file uploads

Marlin.zip

[rondlh]

Do you have a ceramic hotend heater, if so this could be related to #27223

Did you correctly define the MPC_HEATER_POWER?
#define MPC_HEATER_POWER { 40.0f, 40.0f }

You have defined 2 different types of thermistors for the hotends, make sure it's correct.
#define TEMP_SENSOR_0 1010
#define TEMP_SENSOR_1 5

[Drzet]

Do you have a ceramic hotend heater

It's e3d Revo, I believe it's ceramic.

Did you correctly define the MPC_HEATER_POWER?

40W, as per published e3d specs

You have defined 2 different types of thermistors

Yes, the other hotend is the old V6 with pt1000

[rondlh]

Do you have a ceramic hotend heater

It's e3d Revo, I believe it's ceramic.

Did you correctly define the MPC_HEATER_POWER?

40W, as per published e3d specs

You have defined 2 different types of thermistors

Yes, the other hotend is the old V6 with pt1000

OK, good to know.
Here you can see the heater characteristics:
https://e3d-beta.myshopify.com/pages/heatercore-range-datasheet

I think you should look at the orange line, which shows that if you print at 225 degrees C, that the power goes down to about 28W, you could try that number instead of the 40W.

[Drzet]

here's the result @28w
Recv: MPC Autotune finished! Put the constants below into Configuration.h
Recv: MPC_BLOCK_HEAT_CAPACITY 7.49
Recv: MPC_SENSOR_RESPONSIVENESS -1.6147
Recv: MPC_AMBIENT_XFER_COEFF 0.0376
Recv: MPC_AMBIENT_XFER_COEFF_FAN255 0.0699

[rondlh]

All numbers reported seem reasonable, except this minus sign. Are you sure the thermistor type is correct?
Do the readings at room temperature make sense? (Should be about within 1 or 2 degrees of what your bed is reporting).

I'm not sure how you set the new heater power, best way to test is just to issue a M306 P28 before the M306 T. If you change the value in the configuration.h, the new value will not be used immediately... Run an M306 without parameters to see the currently active MPC values.

[Drzet]

Are you sure the thermistor type is correct?

Type 5, as per specifications

Do the readings at room temperature make sense?

B:23 T0:24 T1:23

Run an M306 without parameters

M306 E1 P28.00 C9.36 R0.1810 A0.0525 F0.0884 H0.0036
I changed the R value manually because the negative one was causing thermal runaways.

[tombrazier]

Revo hotends (and PTC hotends generally) break an assumption of the asymptotic MPC autotuning method and this results in nonsense negative sensor responsiveness values.

@StevilKnevil added the differential autotuning method which, I believe, was merged in #25503 (but originally submitted as #25517). This method is slightly less accurate but should give sensible results for all hotend types.

Since #25503 M306 T with no S parameter should attempt the asymptotic method and if that gives a negative sensor responsiveness it then should use the differential method. This bug report perhaps suggests that this default behaviour is not working properly.

@StevilKnevil any thoughts?

[Drzet]

@StevilKnevil added the differential autotuning method...

Yes, I discovered that in Marlin documentation and tried it with the explicit M306 TD command. Got something like -5.5 responsiveness

[tombrazier]

You need M306 T S1 to do a straight differential tune.

[Drzet]

Automatic Differential Tuning
If asymptotic tuning (above) fails, or Differential Tuning is forced with M306 TD then automatic tuning uses the following algorithm.

That's from MPC docs
I'll try S1 later

[tombrazier]

Hmm, looks like the "Automatic Differential Tuning" part of the docs has a few errors. I also see some math values that don't have enough $ signs and are not rendered correctly on my web browser.

But also, the next section in the doc explains why you are still getting negative sensor responsiveness. The value calculated by the differential tuning gets thrown away and replaced with the value from the asymptotic tune. Reading the source code, it does appear to have the described behaviour for the Auto Select tuning case, i.e. M306 T or M306 T S0. When differential tuning is selected explicitly with M306 T S1 I think the differential tuning result will be retained.

[Drzet]

Send: M306 T S1
Recv: MPC Autotune start for E0
Recv: Active Extruder: 0
[...]
Recv: Active Extruder: 0
Recv: X:106.00 Y:110.00 Z:10.00 E:121.71 Count A:52160B:-18880 Z:4000
[...]
Recv: Cooling to ambient
[...]
Recv: Heating to over 200C
[...]
Recv: Measuring ambient heatloss at 209.15
[...]
Recv: MPC Autotune finished! Put the constants below into Configuration.h
Recv: MPC_BLOCK_HEAT_CAPACITY 7.45
Recv: MPC_SENSOR_RESPONSIVENESS -2.4951
Recv: MPC_AMBIENT_XFER_COEFF 0.0427
Recv: MPC_AMBIENT_XFER_COEFF_FAN255 0.0699
[...]
Recv: ok P15 B3

Comparing these values to previous run with M306 TD: M306 E1 P28.00 C9.36 R0.1810 A0.0525 F0.0884 H0.0036 seems like MPC is producing random numbers.

P.S. E0 and E1 are now identical Revo 6

[tombrazier]

Could you include the text that was replaced by [...]?

Also: the values this tune gave are all about 80% of the previous values. Did you use the same power value for both tunes? Marlin has no way of knowing whether the power value is correct so it accepts it as given when tuning and bases all other values on it.

[Drzet]

I have attached logs of M306 T and M306 T S1. Both done with freshly complied latest Marlin 2.1.2.4, right after boot.
E0_M306_T.txt
E0_M306_T_S1.txt

[thisiskeithb]

Both done with freshly complied latest Marlin 2.1.2.4

Testing should be done with bugfix-2.1.x, not a tagged release since releases are actually more than a year behind current code.

[Drzet]

M306 T, M306 T S1 and M306 TD.
Marlin BugFix 02010300
Each tuning done after powering off and on. Same extruder, same power setting of 28W

BUGFIX_E0_M306_T.txt
BUGFIX_E0_M306_T_S1.txt
E0_M306_TD.txt

[tombrazier]

Okay, so...

• M306 TD is incorrect and the documentation needs correcting.
• M306 S1 works and gives very plausible results (for comparison my CHC hotend has MPC_BLOCK_HEAT_CAPACITY 5.54, MPC_SENSOR_RESPONSIVENESS 0.229, MPC_AMBIENT_XFER_COEFF 0.056 and MPC_AMBIENT_XFER_COEFF_FAN255 0.07 which are all very much in the same ballpark as the S1 results for a fairly similar hotend).
• Using the automatic method, i.e. no S parameter to M306, the fallback to differential is incorrectly overridden at the last minute in the "Refining estimates" step. If the automatic method has established that the asymptotic method is incorrect then the refinement step is also incorrect and should not happen.

I can fix both code and docs pretty easily.

[Drzet]

While at it, can you also fix
#if defined(NEOPIXEL_TYPE) && NEOPIXEL_TYPE != NEO_RGB #error "Your FYSETC/MKS/BTT/BEEZ Mini Panel requires NEOPIXEL_TYPE to be NEO_RGB."
in src/inc/SanityCheck.h ?
These come in all sorts, mine is BRG
Allegedly fixed as mentioned here but it actually isn't

[thisiskeithb]

These come in all sorts, mine is BRG

They are all RGB. The fact that you have to set it as BRG is a workaround for a bug on certain motherboards / processors:

• [BUG] BTT Mini 12864 V1 incorrect colors on some STM32F4 & STM32H7 boards #26260

[tombrazier]

It turns out there are several other bugs with the S1 tuning. Fixes for all of them coming soon to a PR near you.